PKU-SEC-Lab / mpcvit

Code release for MPCViT accepted by ICCV 2023
13 stars 1 forks source link

是否有密文上执行推断的开源代码 #5

Open TjFish opened 8 months ago

zwxandy commented 8 months ago

Hi, since this project was completed in collaboration with the company, the Jax code is not stored on our local machine and therefore cannot be released. Thanks for your understanding and if you have any questions or need assistance, we would be happy to provide the guidance. Here are some details about the implementation of secure inference: We implemented the ViT model with SecretFlow framework (SPU Version: 0.3.1b0) in Jax to evaluate the inference latency. We used the semi-honest protocol SEMI-2K, and evaluated the latency under the WAN setting. Our experiments were evaluated on an Intel Xeon CPU@2.40 GHz. We used the PyTorch library for training the ViTs, and reported the model accuracy.

TjFish commented 8 months ago

感谢大佬回复,目前我看原始论文,只有图而没有精确的数据。 不知道大佬能否提供在Cifar10数据集上,MPCVIT各个模型(变种)执行推断的准确率、延迟、通讯开销的准确数据呢,非常感谢!

TjFish commented 8 months ago

例如能否提供原论文中Figure10中关于Cifar10的精确的准确率、延迟、以及通讯开销? Figure 10. Comparison between MPCViT and MPCViT+

yyq-GitHub commented 1 week ago

例如能否提供原论文中Figure10中关于Cifar10的精确的准确率、延迟、以及通讯开销? Figure 10. Comparison between MPCViT and MPCViT+

我想找这个论文数据,没有准确的数据,代码加密部分又无法复现,相对比都难。